Technical Considerations for EndtoEndConversion of Legacy Data to SDTM and ADaM

1
Technical Considerations for End-to-End Conversion
of Legacy Data to SDTM and ADaM

• Prepared for 2007 CDISC Interchange

2
Presentation Outline

• Pilot Objectives
• Technical Processes Used
• Creation of SDTM
• Creation of ADaM
• Validation
• Take-Aways

3
Genentech CDISC Pilot

• Preparing for implementation of CDISC
• Objectives of Pilot were to
• Convert data from a single study to SDTM
• Derive ADaM data with results comparable to
  original analysis datasets
• Investigate various data pathways and methods for
  conducting data conversion
• Bring knowledge about mechanics of conversion
  in-house
• Collect business and process metrics to inform
  the strategic, budgetary, staffing, and technical
  planning for an internal implementation of CDISC
  standards

4
Genentech CDISC Pilot

• We had some CDISC experience
• 1 prior pilot focused only on SDTM
• 1 active ADaM participant
• Others with some CDISC team work
• Limited availability of internal resources
• Hired PharmaStat to provide us
• Mechanics of data conversion
• More CDISC and FDA expertise
• Helpful advice/consultations

5
Primary Pilot Staff

• Genentech
• Amy Black
• Patty Gerend
• Sandra Minjoe
• Peggy Wooster
• Michelle Zhang
• Li Zheng

• PharmaStat
• John Brega
• Jane Diefenbach

Co-authors of this presentation
6
Standards Used

• SDTM IG v3.1.1
• ADaM v2.0
• CRT-DDS v1.0 (define.xml)

7
(No Transcript)
8
Creation of SDTM
9
Creation of SDTM

• All processes were spec-driven
• Source data was converted to structural SDTM
• Controlled terminology was applied as a separate
  step
• Automated tool converted mapping specs into SAS
  code
• Virtually no programming was involved

10
SDTM Findings

• We started with internal raw data
• Operational data made more usable for people
• Combined context and detail
• Added stratification variables
• A few derived variables, like study day
• None of this was helpful for creating SDTM
• Lesson Learned
• Start with original operational data/extracts

11
Creation of ADaM
12
Creation of ADaMTwo Possibilities Considered

• Derive analysis fields directly from SDTM
• Provides traceability
• Logical sequence
• Pass analysis fields from legacy files through
  SUPPQUAL and describe in metadata how it would be
  done from SDTM
• Could be faster
• Concern what we did ? what we described
• Decided to try both and compare

13
ADaM Findings

• Two simple analysis files (ADAE and ADLB) were
  tried using SUPPQUAL to pass through analysis
  variables
• No time savings seen
• Added complexity to specs and review
• Lesson Learned
• Direct SDTM -gt ADaM derivations is the way to go

14
Validation Steps
15
Validation Steps

• Three levels of data checking
• SDTM against Legacy data
• ADaM against Legacy data
• Tables generated from SDTM and ADaM against
  Legacy output
• Used standard templates in data reviews
• Also did full specs reviews
• Most issues discovered were due to unclear specs
  (different interpretations)

16
Validation Findings

• Pilot did not produce a full set of ADaM
  deliverables
• Difficult to create tables directly from SDTM
• Lesson Learned
• Create an ADaM dataset if you need to program any
  tables from the data

17
Other Learnings

• Good specs are essential
• Study expertise is needed when specs are unclear
• Design process flow so specs become your
  metatdata
• Creating define and submittable metadata required
  minor work compared to writing specs
• More bang for the buck
• Develop internal CDISC expertise
• Dont just throw it over the wall to a vendor
• We involved many people at different levels

18
Other Learnings

• Consider using a tool to convert from Legacy -gt
  SDTM
• First time is always most work, but can build
  from there
• With good specs, may not need SAS or study
  experts for this part of the process
• Example
• We already have Informatica (ETL tool) licensed
  in-house for other departments and are testing
  that for possible SDTM mapping work

19
Other Learnings

• Legacy conversion was more resource-intensive
  than we expected
• Traceability
• Consider implementing CDISC structures into
  production rather than relying on legacy
  conversions

20
Technical Considerations for End-to-End Conversion
of Legacy Data to SDTM and ADaM

• Summary
• Bring in any external expertise needed
• Use operational data/extracts -gt SDTM
• Develop internal CDISC expertise
• Consider a tool for SDTM creation
• Use SDTM -gt ADaM
• Derive ADaM dataset for any tables
• Develop specs to be metadata-ready
• Implement CDISC in-stream

21
Thank You

• Contact info
• Sandra Minjoe
• Principal Statistical Programmer Analyst
• sminjoe_at_gene.com
• Questions?